Petr Gusev

Results 38 comments of Petr Gusev
trafficstars

[yes](https://jenkins.scylladb.com/job/sct-github-PRs-scan/job/scylla-dtest/job/PR-2810/lastBuild/testReport/concurrent_schema_changes_test/TestConcurrentSchemaChanges/test_changes_while_node_toggle_raft_/)

For the `io_fiber` [stepdown code](https://github.com/scylladb/scylladb/blob/master/raft/server.cc#L903) to work, both the `joint` and the final `non-joint` configuration must pass through `fsm`, this ensures that `_non_joint_conf_commit_promise` [will be set](https://github.com/scylladb/scylladb/blob/master/raft/server.cc#L883) at that point. This...

the warning problem is fixed [here](https://github.com/scylladb/scylladb/pull/13263), schema commitlog is moved to another directory

the plan: * global `std::optional op_timeout;` in `server::configuration`; * some helper `abort_source_with_timeout` class, which takes `abort_source` and timeout in constructor, provides the combined `abort_source` through `get_abort_source -> abort_source*`, implement it...

A question about `abort_source`. `abort_source` used to communicate aborted state through `abort_requested_exception` -- the `abort_source::request_abort` creates an instance of it with no arguments and `abort_source` clients get this exception when...

> Having only one type of abort thrown by a library makes life of its users easier. But there are disadvantages as well. What are these disadvantages? What value `raft::request_aborted`...

> To propagate the proper reason of the abort to the caller of start_operation we would need to wrap it with try/catch and check the timer again - if it's...

@avikivity can you please give your opinion? `abort_source` doesn't preserve the reason of the abort, if there are many places in code which call `request_abort()` the caller of the aborted...

> > there is no method to extract the exception from the abort_source, only through subscriptions; > > There is --- `check()` I mean direct access to exception_ptr without rethrow,...

> @avikivity can you please give your opinion? > > `abort_source` doesn't preserve the reason of the abort, if there are many places in code which call `request_abort()` the caller...